UB-ECM-PF 91/8

October 1991
arXiv:hep-th/9111015v1 7 Nov 1991

DIFFERENTIAL RENORMALIZATION OF THE

WESS-ZUMINO MODEL ∗

Peter E. Haagensen

Departament d’Estructura i Constituents de la Matèria

Facultat de Fı́sica, Universitat de Barcelona
Diagonal 647, E-08028 Barcelona

Abstract
We apply the recently developed method of differential renormalization to the Wess-
Zumino model. From the explicit calculation of a finite, renormalized effective action,
the β-function is computed to three loops and is found to agree with previous existing
results. As a further, nontrivial check of the method, the Callan-Symanzik equations
are also verified to that loop order. Finally, we argue that differential renormalization
presents advantages over other superspace renormalization methods, in that it avoids
both the ambiguities inherent to supersymmetric regularization by dimensional reduc-
tion (SRDR), and the complications of virtually all other supersymmetric regulators.

∗
Work supported by Ministerio de Educación y Ciencia, Madrid, Spain.
 1. Introduction – Since it has been developed, differential renormalization has had
some important successes. At considerable calculational (and conceptual) simplifica-
tion over other regularization methods, it has been used to fully renormalize massless
λϕ4 theory to three loops[1], to calculate the triangle anomaly[1,2], and the 1-loop
β-function in Yang-Mills theory[1]. Further, mass effects in λϕ4 theory have also been
included straightforwardly[1,2], and the translation of the method to a setting in which
counterterms appear explicitly has also been achieved in the course of proving the
unitarity of the method for massless λϕ4 theory[3]. Like for the λϕ4 model, the per-
turbation theory of the Wess-Zumino model is fairly simple and well-known, but at the
same time sufficiently nontrivial for it to represent a good testing ground for differential
renormalization, and we find that the method works as well in this case as it does in
λϕ4 theory.

The basic idea of the method is as follows: rather than regularizing by altering
the field content or the propagators of a theory (e.g., Pauli-Villars, higher derivatives,
point splitting) or the space on which it is defined (e.g., dimensional regularization),
it regularizes and renormalizes only the x-space amplitudes that are too divergent to
have a Fourier transform into momentum space. In practice, this is done by writing
amplitudes in x-space, expressing the divergent pieces as derivatives of less singular
terms, and then formally performing partial integrations freely (which in fact will be
precisely the implicit regularization and renormalization that the method performs).
Counterterms do not appear at any point (although it is possible to keep track of
them explicitly[3]), and one goes directly from bare amplitudes to renormalized ones,
without a separate intermediate step of subtraction in the regularized theory. The sim-
plest example occurs at one loop both for the 4-point function in λϕ4 theory and the
2-point function in the Wess-Zumino model: these amplitudes will contain the term
((x − x′ )2 )−2 , and its Fourier transform into momentum space (of course, in the end
one must go to momentum space to calculate scattering amplitudes) is:

eip·x 4π 2 ∞ 1
Z Z
4
dx 4 = dr J1 (pr) ,
x p 0 r2

1
where r = (x2 ) 2 , and J1 is the first-order Bessel function. For r → ∞ the integrand
converges fast enough, but for r → 0, J1 (pr) ∼ 12 pr, and we have a logarithmic UV
divergence. To regularize it, differential renormalization then prescribes the use of the
following identity:

1 1 ln x2 M 2
4
=− ✷ , x 6= 0 (1)
x 4 x2

2
valid everywhere except at the origin. Now, the term on which the D’Alembertian is
acting is, as opposed to x14 , a well-defined distribution with a proper Fourier trans-
form. By formally integrating by parts we are then able to define a regularized Fourier
transform for x14 :

1 1 ln x2 M 2 p2 ln x2 M 2
Z Z Z
d4x eip·x = − d4x eip·x ✷ ≡ d4x eip·x
x4 4 x 2 4 x2
p2
= −π 2 ln ,
M̄ 2
where M̄ = 2M γE , and γE = 1.78107... is the Euler-Mascheroni constant. A general fea-
ture of differential renormalization which can already be seen from the example above
is that in the process of expressing singular functions as derivatives of less singular func-
tions and then integrating by parts, thereby discarding (infinite) surface terms, we are,
on the one hand, removing the divergence of an amplitude and, on the other, acquiring
a new mass parameter M which appears in the guise of an integration constant. Since
the method does not keep track explicitly of the infinities being subtracted, this mass
parameter will be crucial for the study of the RG invariance of renormalized amplitudes
and the consistency of the method.

Because Eq.(1) is an identity except at the origin, bare and renormalized ampli-
tudes will also be identical except at isolated points. The method does not, so to speak,
uniformly and blindly regularize an entire theory ab initio (like the regularization pro-
cedures mentioned above), but rather removes divergences singly where they appear in
amplitudes. In this sense, differential renormalization is a “minimal” regularization: it
alters the theory in the least amount possible, and in particular it does not disturb its
original symmetries.

To apply it to the Wess-Zumino model, we will use the manifestly supersymmetric

Feynman rules of Grisaru, Roček, and Siegel[4,5] (except for the fact that we do not go
into momentum space), to calculate loop corrections to the chiral propagator. For su-
perspace theories in general, using these Feynman rules, one ends up with the following
superspace effective action:

Z XZ
Γ= d4θ d4x1 ...d4xN F1 (x1 , θ)F2 (x2 , θ)...FN (xN , θ)G(N ) (x1 , ..., xN ) , (2)
N

where the Fi (xi , θ) are functions of superfields and their (supersymmetry) covariant
derivatives, and G(i) are translationally invariant functions of i spacetime coordinates.

3
Because of the non-renormalization theorem[4,5], the Wess-Zumino model, in particu-
lar, does not have a genuine coupling constant renormalization (but only the one inher-
ited from wavefunction renormalization)[6,7], and is entirely renormalized by removing
the divergences from G(2) (x − x′ ). That we do, up to three loops, with differential
renormalization; with more ease than previous methods, we find a three-loop β-function
which agrees with the existing results[7,8,9], and as further consistency checks of the
(2)
method, we verify the Callan-Symanzik equation for Gren , at each loop order, also to
three loops.

2. The Model and Loop Calculations - The Wess-Zumino model is described by the
following superspace action:

g
Z Z Z Z 
4 2 2 2 3 2 3
S= dx d θd θ̄φφ̄ − d θφ + d θ̄φ̄ , (3)
3!

where (xa , θ α , θ̄ α̇ ), a = 1, 2, 3, 4, α = +, −, α̇ = +̇, −̇ are coordinates of d = 4, N = 1

superspace, and φ (φ̄) is a chiral (antichiral) superfield:

∂
 
D̄α̇ φ = + iθ β σβaα̇ ∂a φ(x, θ, θ̄) = 0
∂ θ̄ α̇
∂
 
a β̇
Dα φ̄ = + iσαβ̇ θ̄ ∂a φ̄(x, θ, θ̄) = 0 .
∂θ α
In component form,
i 1
φ(x, θ, θ̄) = ϕ(x) + θψ(x) + θθF (x) + iθσ a θ̄∂a ϕ(x) − θθ∂a ψ(x)σ a θ̄ + θθ θ̄θ̄✷ϕ(x)
2 4
and
i g 1
Z  
S= d4x −∂a ϕ∗ ∂ a ϕ − (ψσ a ∂a ψ̄ − ∂a ψσ a ψ̄) + F F ∗ − (ϕϕF + ψψA + c.c.) .
4 3! 2
For our calculations in superspace, we use the Feynman rules of Grisaru, Roček and
Siegel[4,5]; in coordinate space, they are: i) the propagator is

1
< φ̄(x, θ)φ(x′ , θ ′ ) > = δ(4) (θ − θ ′ )
4π 2 (x − x′ )2
1
= (θ − θ ′ )2 (θ̄ − θ̄ ′ )2 ;
4π 2 (x − x′ )2
g
ii) for each vertex, include a factor of 3! ; iii) for a chiral (φ3 ) vertex, include two factors
1 2
of − 4 D̄ acting on any two of the propagators for the lines arriving at the vertex; for

4
vertices containing an external line, include only one − 41 D̄ 2 factor acting on any of the
two other (internal) lines; iv) same for antichiral vertices (φ̄3 ), with − 41 D 2 factors; v)
integrate over internal superspace points (xint , θint ) and external θ’s, θext ; vi) include
appropriate symmetry factors for each diagram. Finally, the D-algebra and transfer
rules are:

D 2 D̄ 2 D 2 = 16✷D 2 , D̄ 2 D 2 D̄ 2 = 16✷D̄ 2 ,
D 3 = D̄ 3 = 0 , [Dα , D̄ 2 ] = 4iσαa β̇ ∂a D̄ β̇ ,
Dα (θ ′ , ∂x′ )[δ(4) (θ − θ ′ )f (x − x′ )] = −Dα (θ, ∂x )[δ(4) (θ − θ ′ )f (x − x′ )] .
With these rules for calculating Feynman diagrams, the effective action takes the
form (after all θ-integrations but the last have been carried out):

Z Z 
4 4 4 ′ ′ (2) ′
Γ[φ, φ̄] = dθ d xd x φ(x, θ, θ̄)φ̄(x , θ, θ̄)G (x − x )
Z Z 
2 4 4 4 (3)
+ dθ d x1 d x2 d x3 φ(x1 , θ, θ̄)φ(x2 , θ, θ̄)φ(x3 , θ, θ̄)Gch (x1 , x2 , x3 ) + c.c.
Z
+ d4θ [· · · ] , (4)

where

G(2) (x − x′ ) = δ(4) (x − x′ ) + κ1 (x − x′ ) + κ2 (x − x′ ) + κ3 (x − x′ ) + etc. (5)

(3) g h (4) (4)

i
Gch (x1 , x2 , x3 ) = − δ (x 1 − x 2 )δ (x 1 − x 3 ) + 5 perms. (6)
3!2
and the dots in Eq.(4) refer to higher-point functions which are either finite or are
renormalized by the renormalization of G(2) . The diagrams contributing to G(2) are
depicted in Figs.(1) and (2). κ1 is the one-loop contribution from Fig.(1.a), κ2 the two-
loop contribution, Fig.(1.b), and κ3 contains contributions from the four three-loop
diagrams in Fig.(2).

The non-renormalization theorem determines that the renormalization of g comes

(3)
solely from wavefunction renormalization, and dictates the above form for Gch , correct
(3)
to all loops. By studying the Callan-Symanzik equation for Gch , we find a simple
relaton between the β-function and the anomalous dimension γ(g) of φ:

∂ ∂
 
(3)
M + β(g) − 3γ(g) Gch (x1 , x2 , x3 ; g) =
∂M ∂g
∂
 
(3)
= β(g) − 3γ(g) Gch (x1 , x2 , x3 ; g) = 0
∂g

5
 ⇒ β(g) = 3gγ(g) . (7)
The Callan-Symanzik equation for the renormalized 2-point function then becomes:

∂ ∂
 
M + β(g) − 2γ(g) G(2) ′
ren (x − x ; g, M ) =
∂M ∂g
∂ ∂ 2
  
M + β(g) − G(2) ′
ren (x − x ; g, M ) = 0 . (8)
∂M ∂g 3g
In general, M is a renormalization mass scale that governs the RG flow of the the-
ory; in differential renormalization this scale appears through the process described in
the introduction.

We now proceed to renormalize the theory. For the diagrams of Figs.(1) and (2)
the D-algebra and θ-integrations are straightforward, if tedious, and we present only
the final x-space expressions to be renormalized:

1 g 2 1
 
′
κ1 (x − x ) = 2
(9)
2 4π (x − x′ )4
1 g 4 1 1 1
  Z
κ2 (x − x′ ) = − d4x1 (10)
2 4π 2 (x − x′ )2 (x − x1 )2 (x1 − x′ )4
g 6
2
1 1 1
  Z
4
κ3a (x − x′ ) = d x 1 (11)
8 4π 2 (x − x1 )4 (x1 − x′ )2
1 g 6 1
 
κ3b (x − x′ ) = 2
×
2 4π (x − x′ )2
1 1 1 1
Z Z
d4x1 d4x2 (12)
(x − x1 ) (x1 − x2 ) (x − x2 ) (x2 − x′ )2
4 2 2

1 g 6 1
 
κ3c (x − x′ ) = 2
×
4 4π (x − x′ )2
1 1 1
Z Z
d4x1 d4x2 (13)
(x − x1 ) (x1 − x2 ) (x2 − x′ )4
2 4

1 g 6 1
 
κ3d (x − x′ ) = 2
×
2 4π (x − x′ )2
1 1 1 1 1
Z Z
d4x1 d4x2
(x − x1 ) (x − x2 ) (x1 − x2 ) (x − x1 ) (x − x2 )2
2 ′ 2 2 ′ 2
6
1 g 1

= 6π 4 ζ(3) (14)
2 4π 2 (x − x′ )4

This last (finite) integral was done with the aid of the Gegenbauer polynomial
technique [1,7]. Now, in the course of performing the above integrals, we use the

6
following differential renormalization identities[1] whenever singular terms appear (i.e.,
terms like the l.h.s. below):

1 1 ln x2 M 2
= − ✷ , x 6= 0 (15)
x4 4 x2
ln x M 2
2 1 ln2 x2 M 2 + 2 ln x2 M 2
= − ✷ , x 6= 0 (16)
x4 8 x2
2 2
ln x M 2 1 ln3 x2 M 2 + 3 ln2 x2 M 2 + 6 ln x2 M 2
= − ✷ , x 6= 0 . (17)
x4 12 x2

With this, we are able to calculate the renormalized values of κi :

4
1 ln x2 M 2

2
κren
1 (x) = −2g ✷ (18)
4π x2
 6 2 2
1 ln x M 2 + 2 ln x2 M 2
κren
2 (x) = g4 ✷ (19)
4π x2
ln3 x2 M 2 + 3 ln2 x2 M 2 + 6 ln x2 M 2
 8
1 1
κren
3a (x) = − g6 ✷ (20)
6 4π x2
 8 1 3 2 2 2 2
1 ln x M + 2 ln x M 2 + 4 ln x2 M 2
κren
3b (x) = −g6 ✷3 (21)
4π x2
ln3 x2 M 2 + 3 ln2 x2 M 2 + 6 ln x2 M 2
 8
1 1
κren
3c (x) = − g6 ✷ (22)
3 4π x2
 8 2 2
1 ln x M
κren
3d (x) = −12ζ(3)g 6 ✷ , (23)
4π x2

where we have set x′ = 0 for simplicity. To calculate the β-function and verify the
(2)
consistency of the method we apply the Callan-Symanzik equation, Eq.(4), to Gren =
δ(4) + κren ren ren
1 + κ2 + κ3 . We write

β(g) = β1 g3 + β2 g5 + β3 g7

and separate the pieces in the Callan-Symanzik equation according to order in g 2 and
the coefficients of the δ-function and the different powers of log. All of these have to
vanish separately. At O(g2 ), we get the one-loop β-function:
2
3 1

β1 g3 = g3 . (24)
2 4π

7
 At O(g 4 ), the coefficient of δ(x) vanishes for the following value of the two-loop
β-function:
4
3 1

β2 g5 = − g5 , (25)
2 4π

2 2
and the coefficient of ✷ ln xx2M vanishes for the above value of β1 ; this is a consistency
check of the method. At O(g 6 ), the vanishing of the coefficient of δ(x) gives the following
three-loop β-function:
6
21 1
 
7
β3 g = + 9ζ(3) g7 . (26)
4 4π

2 2 2 2 2
Finally, the coefficients of ✷ ln xx2M and ✷ ln xx2M will vanish for the above values
of β1 and β2 , and these are then two further consistency checks. Our final result for
the β-function is then:
"  2 4 6 #
3 g 3 g 21 g
 
β(g) = g − + ( + 9ζ(3)) . (27)
2 4π 2 4π 4 4π

We note that although the three-loop part is scheme-dependent, as opposed to

the one- and two-loop results, the coefficient of ζ(3) should be universal (barring, of
course, a coupling constant redefinition involving ζ(3)) because the only diagram at
three loops leading to the transcendental function ζ(3) is primitively divergent. While
our coefficient of ζ(3) coincides with that obtained in [9], it does not agree with [7]. The
correct result is that of [9], however[10]; in fact, because that is a four loop calculation,
it is possible to verify the three-loop result by RG pole equations (which the authors
of [9] in fact do).

Conclusions - We have seen that the method of differential renormalization is read-

ily applicable to the superspace Wess-Zumino model and, in particular, the β-function
calculations were considerably simpler than in SRDR [7,9]. Firstly, the integrations
performed here were either trivial (integrations with δ-functions) or very simple con-
volutions; secondly, we did not have to keep track of subtraction of subdivergences: at
each loop order the renormalization is done in a single step, by the direct use of “dif-
ferential renormalization identities” like Eqs.(12)-(14) whenever singular expressions
are encountered (in SRDR, on the other hand (cf. Eq.(2.11) of [7]), a careful account
needs to be kept of different renormalization constants corresponding to contributions

8
of subdivergences at higher loops).

We have attempted to describe the differential renormalization of the Wess-Zumino

model in such a way that the extension to other superspace theories becomes obvi-
ous: one should again start with the general form for the effective action, Eq.(2), with
the G(N ) written in terms of x-space integrals, and apply differential renormalization
identities to these integrals, thus regularizing and renormalizing the theory. Differ-
ential renormalization will take translation-invariant quantities G(N ) into translation-
(N )
invariant quantities Gren , and will manifestly maintain supersymmetry. Naturally, we
expect this procedure to be far simpler than several other existing regularization meth-
ods (e.g., the supersymmetric versions of Pauli-Villars, higher derivative and point-
splitting methods). Furthermore, if the method proves to be as simple to implement
in general as SRDR, it would have the advantage that the ambiguities associated with
SRDR would be avoided: like for dimensional regularization, SRDR presents ambigu-
ities due to the presence of intrinsically four-dimensional quantities, like εµνρσ , in a
dimensionally continued setting [5,11]. Differential renormalization, of course, never
leaves four dimensions and would thus avoid these problems altogether. One initial
indication of the further applicability of the method is the fact that gauge (vector)
superfield propagators are identical to the massless chiral propagators we have consid-
ered here. Thus, calculations with vector superfields should not in principle present
any new difficulty; for instance, the one-loop correction to the vector propagator in
supersymmetric Yang-Mills theory coupled to a massless chiral superfield is again as
simple to compute as in SRDR. Further work along these lines is in progress.

Acknowledgments - It is a pleasure to thank J.I. Latorre for always illuminating

discussions. Comments on the manuscript from D.Z. Freedman and J.I. Latorre are
also very appreciated.

References

[1] Freedman, D.Z., K. Johnson and J.I. Latorre, to be published in Nucl.Phys. B.

[2] Freedman, D.Z., Proceedings of the Stony Brook Conference on Strings and
Symmetries, Spring 1991.
[3] Freedman, D.Z., R. Muñoz-Tapia, X. Vilası́s-Cardona, manuscript in preparation.
[4] Grisaru, M.T., M. Roček and W. Siegel, Nucl.Phys. B159(1979)429.
[5] Gates, J., M.T. Grisaru, M. Roček and W. Siegel, Superspace, Benjamin-Cummings,
1983.
[6] Iliopoulos, J. and B. Zumino, Nucl.Phys. B76(1974)310.
Ferrara, S., J. Iliopoulos and B. Zumino, Nucl.Phys. B77(1974)413.

9
[7] Abbott, L.F. and M.T. Grisaru, Nucl.Phys. B169(1980)415.
[8] Townsend, P.K. and P. van Nieuwenhuizen, Phys.Rev. D20(1979)1832.
[9] Sen, A. and M.K. Sundaresan, Phys.Lett. B101(1981)61.
[10] M.T. Grisaru, private communication.
[11] Siegel, W., Phys.Lett. B94(1980)37.

Figures

Fig.(1): One- and two-loop contributions to G(2) (x − x′ ).

Fig.(2): Three-loop contributions to G(2) (x − x′ ).

10